# Assessing Shape Bias Property of Convolutional Neural Networks

arXiv: https://arxiv.org/abs/1803.07739 · HTML: https://mlsanity.com/paper/1803.07739/

## Key facts

- Cited by 18 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 21 Mar 2018.

## Built on by (most cited first, 18 of 18)

- [X-Adv: Physical Adversarial Object Attacks against X-ray Prohibited Item Detection](https://mlsanity.com/paper/2302.09491/) (2023)
- [LucidPPN: Unambiguous Prototypical Parts Network for User-centric Interpretable Computer Vision](https://mlsanity.com/paper/2405.14331/) (2024)
- [The Emergence of the Shape Bias Results from Communicative Efficiency](https://mlsanity.com/paper/2109.06232/) (2021)
- [Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild](https://mlsanity.com/paper/2503.10065/) (2025)
- [Convolutional Neural Networks Rarely Learn Shape for Semantic Segmentation](https://mlsanity.com/paper/2305.06568/) (2023)
- [Understanding Segment Anything Model: SAM is Biased Towards Texture Rather than Shape](https://mlsanity.com/paper/2311.11465/) (2023)
- [Can CLIP Count Stars? An Empirical Study on Quantity Bias in CLIP](https://mlsanity.com/paper/2409.15035/) (2024)
- [Shape-Biased Domain Generalization via Shock Graph Embeddings](https://mlsanity.com/paper/2109.05671/) (2021)
- [A Mathematical Philosophy of Explanations in Mechanistic Interpretability -- The Strange Science Part I.i](https://mlsanity.com/paper/2505.00808/) (2025)
- [Decomposing and Revising What Language Models Generate](https://mlsanity.com/paper/2509.00765/) (2025)
- [GeoPE:A Unified Geometric Positional Embedding for Structured Tensors](https://mlsanity.com/paper/2512.04963/) (2025)
- [Abutting Grating Illusion: Cognitive Challenge to Neural Network Models](https://mlsanity.com/paper/2208.03958/) (2022)
- [Will the Inclusion of Generated Data Amplify Bias Across Generations in Future Image Classification Models?](https://mlsanity.com/paper/2410.10160/) (2024)
- [SSR-GNNs: Stroke-based Sketch Representation with Graph Neural Networks](https://mlsanity.com/paper/2204.13153/) (2022)
- [On the ability of CNNs to extract color invariant intensity based features for image classification](https://mlsanity.com/paper/2307.06500/) (2023)
- [On the Influence of Shape, Texture and Color for Learning Semantic Segmentation](https://mlsanity.com/paper/2410.14878/) (2024)
- [Contrast All the Time: Learning Time Series Representation from Temporal Consistency](https://mlsanity.com/paper/2410.15416/) (2024)
- [ShapeY: A Principled Framework for Measuring Shape Recognition Capacity via Nearest-Neighbor Matching](https://mlsanity.com/paper/2604.25065/) (2026)

## Abstract (arXiv)

It is known that humans display "shape bias" when classifying new items, i.e., they prefer to categorize objects based on their shape rather than color. Convolutional Neural Networks (CNNs) are also designed to take into account the spatial structure of image data. In fact, experiments on image datasets, consisting of triples of a probe image, a shape-match and a color-match, have shown that one-shot learning models display shape bias as well. In this paper, we examine the shape bias property of CNNs. In order to conduct large scale experiments, we propose using the model accuracy on images with reversed brightness as a metric to evaluate the shape bias property. Such images, called negative images, contain objects that have the same shape as original images, but with different colors. Through extensive systematic experiments, we investigate the role of different factors, such as training data, model architecture, initialization and regularization techniques, on the shape bias property of CNNs. We show that it is possible to design different CNNs that achieve similar accuracy on original images, but perform significantly different on negative images, suggesting that CNNs do not intrinsically display shape bias. We then show that CNNs are able to learn and generalize the structures, when the model is properly initialized or data is properly augmented, and if batch normalization is used.